Structure-coevolution dual-guided engineering of C-glycosyltransferase enables high-level biosynthesis of vitexin in Yarrowia lipolytica

TL;DR

Scientists improved an enzyme to efficiently produce a valuable flavonoid in yeast, enabling sustainable biosynthesis.

Contribution

A structure-coevolution dual-guided strategy enhanced C-glycosyltransferase activity and enabled high vitexin production in Yarrowia lipolytica.

Findings

Engineered TcCGT1 variant M7 showed 43.1-fold higher activity toward apigenin and up to 55.0-fold for other substrates.

Metabolic engineering in Yarrowia lipolytica achieved 361.0 mg/L vitexin, a 6.3-fold improvement over wild-type.

Fed-batch fermentation produced 2.14 g/L vitexin, demonstrating industrial scalability.

Abstract

Flavonoid C-glycosides are bioactive compounds with significant pharmaceutical potential. While biosynthesis offers a sustainable and green alternative to traditional chemical synthesis, its industrial scalability has been hindered by the poor catalytic efficiency and low thermostability of natural C-glycosyltransferases (CGTs). In this study, we report a structure-coevolution dual-guided engineering strategy to modify TcCGT1 from Trollius chinensis, a pivotal enzyme in flavonoid C-glycosylation. The engineered variant M7 exhibited a 43.1-fold improvement in catalytic activity toward apigenin and up to 55.0-fold enhancement for other flavonoid acceptors. It also exhibited an 9.2-fold increase in half-life at 30 °C. Molecular dynamics simulations and structural analyses revealed that the mutations remodeled the substrate-binding cavity and optimized its interactions. To translate this enzymatic advance into a biosynthetic platform, we integrated the optimized variant into a de novo synthetic pathway and conducted metabolic engineering in Yarrowia lipolytica. This integration achieved a vitexin titer of 361.0 mg/L, a 6.3-fold improvement over the wild-type-expressing strain. Furthermore, the engineered microbial cell factory produced 2.14 g/L of vitexin in fed-batch fermentation, demonstrating industrial potential. This study provides new prospects for the structure-coevolution dual-guided synergistic framework of CGT engineering and a sustainable synthetic biology platform for high-value flavonoid C-glycosides.

Image 1

Figures

B–S7-S20). Notably, M7 achieved a 43.1-fold increase in activity toward apigenin and a greater than 50.0-fold enhancement for kaempferol. In terms of synthesis capacity, M7 produced 3.69 ± 0.02 g of vitexin per liter of LB medium, which is a 62.1-fold increase over the wild-type ([Fig. 3](#fig3)A). This result demonstrates that the integrated structure-coevolution engineering strategy not only enhanced the specific activity and stability but also substantially improved the total catalytic output of the enzyme.Fig. 3Enhanced catalytic performance and broadened substrate scope of engineered TcCG

Full text

Introduction

1

Flavonoid C-glycosides are bioactive natural products with proven antioxidant, anti-inflammatory, and antiproliferative activities [[1], [2], [3]]. The C–C bond between the aglycone and the glycosyl moiety enhances metabolic stability and pharmacokinetics compared to their O-linked counterparts [4,5]. Among these compounds, vitexin stands out due to its cardioprotective, neuroprotective, and antidiabetic effects [6]. Flavonoid C-glycosides are typically extracted from natural sources, which is time-consuming, resource-intensive, and low-yielding [4,7]. Chemical synthesis remains challenging due to multi-step processes, toxic catalysts, and extensive organic solvents, which makes commercial-scale synthesis unsustainable and environmentally harmful [8,9]. These constraints highlight the need for efficient, sustainable biosynthesis of flavonoid glycosides in microbial hosts using inexpensive resources [[10], [11], [12]].

C-glycosyltransferases (CGTs) are the key enzymes that catalyze the C-glycosylation reactions by the formation of C-glycosidic bonds in plants [13]. Recently, biotechnological production of C-glycosides has been driven by the discovery of CGTs [4,5]. However, natural CGTs suffer from low catalytic efficiency, expression issues, and limited resources, preventing them from meeting industrial biosynthesis requirements [14,15]. Therefore, overcoming the limitations of CGTs is essential for unlocking the entire bioconversion pathway to high-value C-glycosides. To address these bottlenecks, structure-guided protein evolution strategies, such as combinatorial active-site saturation testing (CAST) and iterative saturation mutagenesis (ISM), and focused rational iterative site-specific mutagenesis (FRISM) have been extensively studied. These methods aim to alter key residues within the active sites, leading their catalytic functions to meet specific requirements [[16], [17], [18]]. Despite current advancements, this approach often triggers the activity-stability trade-off [19] by disrupting the essential balance between catalytic flexibility and global stability [20,21]. To mitigate this trade-off, stepwise solutions have been explored. For instance, structure-guided mutagenesis of UGT708A60 from Hordeum vulgare improved catalytic activity toward phloretin, but compromised thermostability. Researchers subsequently employed the computer-aided stability optimization tool FireProt to restore and enhance its thermal performance [22]. The sequential strategy of “activity first, stability second” treats the two objectives as separate engineering modules.

Recently, focusing on functionally impactful distal residues has emerged as a powerful solution to overcome the limitations of active-site optimization [23], as exemplified by global mutagenesis of dehydrogenase and proline-induced loop engineering in TbSADH [24,25]. The core strategy, coevolution-guided engineering, employs constraints from multiple sequence alignments (MSAs) to identify functionally coupled residue pairs across the protein, including those distant from the catalytic center, enabling the rational design of mutations to regulate enzyme activity and stability [26,27]. Comprehensive platforms like EVcouplings use cross-species evolutionary constraints to identify functionally critical residues and can predict mutation effects with high accuracy [28,29]. These insights are crucial for understanding how mutations affect protein stability, folding, and interactions, particularly for distal sites difficult to capture in crystal structures. Despite its potential to overcome local optimization limits, coevolution-guided engineering has an inherent limitation. It focuses on optimizing distal residues while lacking concurrent refinement of the active sites. This often results in modest improvements in the core catalytic efficiency of enzymes toward native substrates. This gap is pronounced for glycosyltransferases, where coevolution-guided engineering is still in its infancy. Thus, it is critical to develop a novel structure-coevolution dual-guided strategy for the synergistic modification of C-glycosyltransferases in both active sites and distal sites.

Furthermore, it is crutial to establish a green, sustainable bioconversion process centered on microbial synthesis with both engineered enzymes and a suitable cellular chassis. Yarrowia lipolytica (Y. lipolytica) was selected as the ideal cellular chassis due to its ability to sustain high flux toward the essential precursors, acetyl-CoA and malonyl-CoA. Previous studies have successfully reconstructed the pathways in Y. lipolytica, demonstrating its robust metabolic flexibility and genetic tractability [30,31]. In addition, this chassis tolerates high cell density fermentation, aligning with the industrial requirements, and offers advantages over Saccharomyces cerevisiae and Escherichia coli, which are limited by precursor supply and intracellular oxidative stress, respectively [[10], [11], [12]]. The biosynthetic pathway of vitexin involves a series of enzymatic reactions. First, l-phenylalanine/l-tyrosine is deaminated to p-coumaric acid (p-CA) by l-phenylalanine/tyrosine ammonia lyase (TAL/PAL), and then 4-coumaroyl-CoA is produced by 4-coumaroyl-CoA ligase (4CL). Naringenin chalcone is synthesized by chalcone synthase (CHS), combined with three molecules of malonyl-CoA, and finally, naringenin chalcone is catalyzed by chalcone isomerase (CHI) to produce (2S)-naringenin, a core flavonoid intermediate [32]. Flavone synthase (FNS) then introduces a double bond in ring C of naringenin, converting naringenin into apigenin. Subsequently, C-glycosyltransferase (CGT) catalyzes 8-C-glycosylation of apigenin, producing vitexin [33]. The intracellular cell factory can be optimized through a systematic approach, refining three key components: the shikimate pathway for precursor supply, the core flavonoid synthesis pathway, and the glycosylation module anchored by an engineered CGT.

This study aims to establish a full-chain synthetic biology platform encompassing rational design of enzyme elements, optimization of chassis metabolic networks, and scaling-up of fermentation processes. TcCGT1 from Trollius chinensis, the first structurally characterized flavone 8-C-glycosyltransferase, was selected as a model enzyme [34]. We developed a structure-coevolution dual-guided engineering strategy designed to synergize precise active-site optimization with coevolution-informed distal-site mutagenesis. This integrated approach aimed to systematically overcome the activity-stability trade-off that has limited previous engineering efforts. The engineered variant was then integrated into a metabolically engineered Yarrowia lipolytica chassis to construct an efficient and robust cell factory, enabling the high-titer de novo synthesis of vitexin from glucose. This work intends to provide a transferable exemplary case for CGT engineering, and offer a sustainable solution for the green manufacturing of high-value flavonoid C-glycosides.

Materials and methods

2

Enzyme cloning, expression, and purification

2.1

The DNA sequence of TcCGT1 was synthesized by GenScript (China) and inserted into the BamHI-XhoI sites of the pET-28a vector. The pET-28a-TcCGT1 plasmid served as the template for site-directed mutagenesis. All primers used in this study are listed in Table S1. Mutations were introduced using the Mut Express II Fast Mutagenesis Kit V2 (Vazyme, China). The recombinant plasmids were transformed into E. coli BL21(DE3) (Sangon Biotech, China) for heterologous protein expression. Transformed cells were first cultured overnight at 37 °C, shaking at 225 rpm in 10 mL of Luria-Bertani (LB) medium with 50 μg/mL kanamycin. This starter culture was then used to inoculate 200 mL of fresh LB medium containing 50 μg/mL kanamycin. Cells were grown at 37 °C until the optical density at 600 nm (OD_600_) reached 0.6–0.8. Protein expression was induced by adding 0.2 mM isopropyl β-d-1-thiogalactopyranoside (IPTG), followed by incubation for 20 h at 16 °C. Cells were harvested by centrifugation at 8000×g for 10 min at 4 °C. The cell pellet was resuspended in 15 mL of lysis buffer (50 mM sodium phosphate, 500 mM NaCl, and 5 mM imidazole; pH 7.8) and lysed by sonication on ice for 10 min. Cell debris was removed by centrifugation at 12,000×g for 20 min at 4 °C. The soluble supernatant was filtered through a 0.45 μm syringe filter and subsequently purified using nickel-nitrilotriacetic acid (Ni-NTA) affinity chromatography.

Glycosyltransferase activity assay and analytical methods

2.2

The standard enzymatic reaction mixture (200 μL) consisted of 50 mM potassium phosphate buffer (pH 8.0), 1 mM UDP-Glucose (UDP-Glc), 0.5 mM apigenin, 5% (v/v) DMSO, and 1–20 μg purified enzyme. The reaction was incubated at 30 °C and terminated by adding an equal volume (200 μL) of methanol. All assays were performed in triplicate. To investigate the acceptor selectivity of the mutants, reactions were tested with alternative acceptors (compounds 1–15) using UDP-Glc as the sugar donor. The compound information is listed in Table S3. To assess sugar donor selectivity, reactions were performed with apigenin (1) as the acceptor and various donor substrates, including UDP-glucuronic acid (UDP-GlcA), UDP-xylose (UDP-Xyl), UDP-galactose (UDP-Gal), UDP-N-acetylglucosamine (UDP-GlcNAc), UDP-N-acetylglucosamine azide (UDP-GlcNAz), UDP-N-acetylgalactosamine (UDP-GalNAc), and UDP-N-acetylgalactosamine azide (UDP-GalNAz) [35]. The reaction mixtures were centrifuged at 12,000×g for 10 min, and the supernatants were filtered through a 0.22 μm membrane and analyzed by high-performance liquid chromatography (HPLC) or HPLC-mass spectrometry. HPLC was performed on a Shimadzu Prominence LC-20A system equipped with a C18 column (250 mm × 4.6 mm, 5 μm particle size) and an SPD-20A UV-Vis detector. The HPLC methods are provided in Table S2.

Construction of saturated mutation libraries and screening

2.3

Saturated mutagenesis at targeted sites was performed using primers containing NNK degenerate codons. For each site-saturation library, 92 individual colonies were picked and cultured in 96-deep-well plates containing 200 μL of LB medium with 50 μg/mL kanamycin at 37 °C overnight. These cultures were then transferred to 1 mL of fresh LB medium with kanamycin in new deep-well plates. When the OD_600_ reached 0.6–0.8, protein expression was induced with 0.2 mM IPTG at 16 °C for 20 h. Cells were harvested and subjected to two freeze-thaw cycles. The pellets were resuspended in 200 μL of lysis buffer, and cell debris was removed by centrifugation. For high-throughput screening, 10 μL of the crude cell lysate was transferred to a 96-well PCR plate (each sample in triplicate) and added to a 100 μL reaction mixture as described above. A no-enzyme control, containing 10 μL of lysis buffer, was included. The reactions were incubated at 30 °C for 2 h and terminated by heating at 90 °C for 5 min. Precipitated material was removed by centrifugation at 12,000×g for 20 min. Subsequently, 50 μL of the supernatant was transferred to a new well, and an equal volume of ethyl acetate:chloroform (3:1, v/v) was added for liquid-liquid extraction. The mixture was vigorously agitated using an eight-channel pipettor, and the aqueous phase was isolated using a Whatman 1 PS phase separator filter paper (Cytiva, USA). The extraction was repeated once. Finally, 30 μL of the purified aqueous phase was transferred to a 96-well plate and diluted with 30 μL of distilled water and 60 μL of a 0.2% (w/v) 2-aminoethyl diphenylborinate solution in 20% (v/v) ethanol. The absorbance was measured at 394 nm using a microplate reader (BioTek Instruments, USA).

Kinetic parameter determination

2.4

The kinetic parameters of the wild-type (WT) and mutant enzymes (M4, M7) were determined for apigenin and UDP-Glc. For assays with varying apigenin concentrations, the reaction mixture (200 μL) contained 50 mM potassium phosphate buffer (pH 8.0), 1 mM UDP-Glc, 0.25 μg of purified enzyme, and apigenin at concentrations of 5, 10, 20, 50, 75, and 100 μM (for WT) or 5, 10, 20, 50, 75, 100, 150, and 200 μM (for M4 and M7). For assays with varying UDP-Glc concentrations, the mixture contained 0.2 mM apigenin, 1 μg of purified enzyme, and UDP-Glc at concentrations of 5, 10, 20, 50, 75, 100, 150, and 200 μM. All reactions were conducted at 30 °C for 10 min and quenched with an equal volume of methanol. All experiments were performed in triplicate. The Michaelis-Menten constant (Km) and turnover number (kcat) were determined by fitting the data to the Michaelis-Menten equation using nonlinear regression.

Computational methods

2.5

Amino acid conservation analysis was performed using ConSurf Web Server [36]. The complete EVcouplings analysis pipeline, including redundant sequence removal, sequence weighting calculation, and residue-residue coupling prediction, was executed using the EVcouplings toolset (evcouplings.org). The multiple sequence alignment was constructed by iteratively searching the UniRef 90 database. To ensure quality, a standard two-step filtering protocol was applied: sequences were clustered at a 90% pairwise identity threshold to reduce redundancy, followed by the application of a minimum column coverage threshold of 0.70 to filter out fragmented sequences. Homology models of the M4 and M7 mutants were generated using the Swiss-Model server with the TcCGT1 crystal structure as a template. Molecular docking was performed using AutoDock Vina 1.5.6. Structural visualization and analysis were conducted using PyMOL. The spatial properties of the substrate-binding cavity were calculated using CavityPlus 2022 [37]. Molecular dynamics simulations were performed using the AMBER 24 software package.

Y. lipolytica plasmid and strain construction

2.6

All plasmids and Y. lipolytica strains constructed in this study are detailed in Table S4. All plasmids were constructed using Gibson assembly [38]. All heterologous genes were synthesized by GenScript (China). The CRISPR-Cas9 system and multicopy integration tool were consistent with previously reported methods [39,40]. The plasmids were constructed in and amplified from E. coli DH5α. The Y. lipolytica Δ ku70 strain, derived from the Po1f strain (ATCC MYA-2613) by knocking out the ku70 gene with CRISPR-Cas9, was used as the starting strain [38]. Y. lipolytica strains were cultured at 30 °C on yeast extract glucose (YPD) medium or yeast nitrogen base (YNB) medium with uracil or leucine as needed. The sources and descriptions of all genes are provided in Table S5.

Shake-flask and fed-batch fermentation

2.7

For shake-flask fermentations, strains were streaked onto YPD agar plates and incubated at 30 °C for 2–3 days. Single colonies were inoculated into 10 mL of YPD medium in a shake flask and cultured at 30 °C with shaking at 225 rpm for 24 h. This seed culture was then transferred to a new flask containing 50 mL of fresh YPD medium and fermented for 120 h under the same conditions.

For fed-batch fermentation in a bioreactor, the seed culture was transferred to a 5 L bioreactor containing 2.3 L of defined inorganic salt medium. Each liter of medium contained 3.5 g KH_2_PO_4_, 5 g (NH4)2_SO_4, 0.5 g MgSO_4_·7H_2_O, and 40 g d-glucose, 2 g uracil, 2 g leucine, with an additional 3 mL of metal ions and 5 mL of vitamin solution to promote growth [41]. The agitation speed ranged from 300 to 800 rpm, and the dissolved oxygen (DO) level was maintained at 20%–30% through automatic adjustment of stirrer speed and aeration. The pH was maintained at approximately 5.0 using 4 M NaOH. A feed containing 600 g/L glucose was initiated when the residual glucose concentration decreased to 1 g/L. The substrate feeding was performed in a continuous and automated mode. The feeding rate of glucose solution was initially set at 5 mL/h, and the residual glucose concentration was measured every 6 h to adjust the feeding rate for maintaining the residual glucose below 1 g/L. Nitrogen sources were supplemented starting from 60 h. The organic nitrogen supplement contained 40 g/L tryptone and 10 g/L yeast extract. The inorganic nitrogen sources included 13.5 g/L KH_2_PO_4_, 25 g/L (NH_4_)2_SO_4, and 2.5 g/L MgSO_4_·7H_2_O. Additionally, 2 g/L uracil, 2 g/L leucine, and 10 mL of trace metal and vitamin supplement solution were added to both nitrogen source systems. A volume of 900 μL methanol was added to a volume of 100 μL culture broth, and the mixture was vortexed vigorously for 10 min.

Results and discussion

3

To improve the catalytic performance of TcCGT1, we devised a structure-coevolution dual-guided engineering strategy involving the collaborative engineering of the active sites and distal sites. First, semirational evolution of active sites was performed under the guidance of structure-derived information. Potential active-site libraries were identified through alanine scanning, followed by the high-throughput screening based on CAST/ISM. Second, the engineering of distal sites was conducted based on sequence-derived information. Coevolution analysis was used to identify distal hotspots predicted by virtual screening. High-performance mutants were identified through focused rational iterative site-specific mutagenesis (FRISM) [42]. (Scheme 1).Scheme 1Structure-coevolution dual-guided engineering workflow of TcCGT1. The engineering strategy comprises two synergistic phases. Structure-guided engineering: Potential hotspot residues were predicted via molecular docking and multiple sequence alignment. Following alanine scanning, iterative saturation mutagenesis (ISM) coupled with a biphasic high-throughput screening assay yielded the intermediate variant M4. Coevolution-guided engineering: A coevolutionary analysis platform performed virtual saturation mutagenesis on M4. Subsequently, focused rational iterative saturation mutagenesis (FRISM) was conducted to combinatorially assemble beneficial distal mutations, culminating in the final, optimized variant M7.Scheme 1

Structure-guided engineering of active sites by CAST/ISM

3.1

Apigenin and UDP-glucose were both docked into TcCGT1's crystal structure (PDB:6JTD) because we hypothesized that enhancing donor-binding efficiency could generally improve the activity toward diverse acceptors. 29 and 18 residues were identified within 4 Å of apigenin and UDP-glucose, respectively. Multiple sequence alignment (MSA) was conducted to obtain the conservation scores of all residues (Fig. S1). To preserve core catalytic function, highly conserved residues with conservation degree >7 calculated by Consurf website [36] were excluded (Fig. S1), resulting in 17 variable positions selected for alanine scanning to pinpoint the hotspots for CAST (Fig. 1A). The catalytic activity of the alanine-scanning mutants was assessed as shown (Fig. 1B).Fig. 1Structure-guided engineering of TcCGT1 for active sites. (A) Molecular docking of substrates apigenin and UDP-Glc into the active site of TcCGT1 (PDB: 6JTD). Residues within 4 Å are shown. (B) Relative activity of alanine-scanning mutants at selected positions (N.D., not detected). (C) Principle of the biphasic high-throughput screening assay, which exploits the polarity shift upon glycosylation to separate product (aqueous) from unreacted substrate (organic). (D) Validation of the high-throughput screening assay using crude cell extracts, yielding a Z′-factor of 0.843, indicative of a robust assay (Z′-factor >0.5). (E) Activity profile of the saturation mutagenesis library at the key residue L145. (F) Progressive improvement in specific activity over five rounds of iterative saturation mutagenesis (ISM). Numbers denote fold-increase relative to wild-type. M1: L145M, M2: N50S/L145M, M3: N50S/K92T/L145M, M3-1: N50S/K92T/C321T, M4: N50S/K92T/L145M/C321S, M4-1: N50S/K92T/L145M/C321V, M4-2: N50S/K92T/L145M/C213A, M5: N50S/K92T/L145M/C321S/R253A. Error bars in (B) and (E) represent the standard deviation of three replicates. Error bars in (D) represent the standard deviation of eight replicates.Fig. 1

Based on the screening results, four sites including N50, K92, R253, and C321 with higher activity were chosen for the ISM. In contrast, L145A resulted in a complete loss of activity, unequivocally identifying this residue as critical for catalytic function. It is hypothesized that the leucine residue resides in the second coordination sphere, owing to its nonpolar nature and its location within 5 Å of the substrate. This residue likely provides a specific hydrophobic microenvironment for the substrate through hydrophobic interactions and plays a crucial role in stabilizing the local conformation of the active site. Given its strong phenotypic impact and potential, L145 was prioritized for saturation mutagenesis.

A miniaturized biphasic fluorescence-based high-throughput screening assay was developed [43]. In this method, aglycones migrated into an ethyl-acetate overlay and were trapped by phase-selective filter paper, while hydrophilic C-glycosylated products were retained within discrete aqueous droplets (Fig. 1C). Significant differences in fluorescence values were observed between the negative control and the two experimental groups. The Z′-factor for the crude enzyme extract group reached 0.843, meeting the standard for a reliable high-throughput screening assay (Fig. 1D).

As anticipated, variant L145M demonstrated a nearly five-fold increase in activity compared with the wild-type enzyme (WT), outperforming the previously identified N50A from alanine scanning (Fig. 1E). The primary function of this leucine residue in the second coordination sphere is to maintain both the size and hydrophobicity of a crucial hydrophobic cavity. The substitution with alanine disrupts this foundation, resulting in a loss of activity. In contrast, methionine not only fulfills these basic requirements but also possesses a longer and more flexible side chain. This flexibility likely allows for more optimized conformational adjustments of the active site during substrate binding or catalysis. L145M was selected as the template for subsequent iterative saturation mutagenesis. After five iterative rounds of saturation mutagenesis of approximately 1000 variants, variant M4 (N50S/K92T/L145M/C321S) was found, which exhibited a 27.7-fold improvement in catalytic ability over the WT, validating the efficacy of the structure-guided CAST/ISM framework for active-site refinement (Fig. 1F).

Coevolution-based engineering of distal sites by FRISM

3.2

Based on structure-guided mutagenesis of the active site, a coevolution-based strategy was employed to exploit functionally related distal residues. We performed this analysis with the Evcouplings platform, a tool designed to identify key residues and predict mutational effects. The described protocol resulted in a deep and diverse alignment. From an initial pool of 79,398 homologous sequences, the final alignment robustly covered the core domain from residue 14 to 457, spanning 444 positions. Within this region, 80.7% of columns (390 positions) maintained high sequence occupancy. The phylogenetic diversity, quantified by the effective number of sequences (Neff), was 55,555, yielding an Neff/L ratio of approximately 142.4. This exceptionally high Neff value strongly indicates that our coevolutionary signal is derived from a broad and diverse evolutionary landscape, thereby significantly mitigating concerns of sequence redundancy, phylogenetic bias, or spurious correlations. Virtual saturation mutagenesis was performed at approximately 400 amino acid sites, generating and computationally evaluating about 8000 mutant variants. Coevolutionary constraints from phylogenetically diverse multiple sequence alignments were quantified to calculate statistical energy changes (ΔE), enabling reliable phenotypic prediction. Based on ΔE, mutations were classified as deleterious, neutral, or beneficial and ranked by their effect strength scores. To systematically prioritize candidates, we first focused on distal regulatory regions, identifying sites enriched with beneficial mutations, as visualized in a screening heatmap (Fig. 2A). Subsequently, stringent effect strength score thresholds were applied to select the most promising single-point mutations from these regions, yielding six high-impact mutations (score >4) and eight moderate-impact mutations (score 3–4). Finally, to explore potential synergy and minimize epistatic interference, spatially adjacent high-scoring residues (pairs 134/138 and 450/451) were combinatorially engineered into double mutants. All rationally designed variants were then incorporated into the M4 protein backbone through two rounds of FRISM, as our engineering strategy aimed to iteratively optimize this already improved scaffold (Fig. 2B).Fig. 2Coevolution-guided engineering of distal sites. (A) Virtual saturation mutagenesis predicts beneficial mutations. The heatmap displays effect strength scores from an EVcouplings-based analysis; mutations with scores >4 and 3–4 are boxed with solid and dashed lines, respectively. (B) Focused rational iterative saturation mutagenesis (FRISM) experimentally validates and combines top-predicted mutations, yielding the final M7 variant. Mutations introduced in the sixth and seventh rounds are highlighted in blue and yellow, respectively. Error bars represent the standard deviation of three replicates.Fig. 2

By screening only 16 combinatorial mutants, we identified two with significantly improved activity: the first-round mutant M6 (N50S/K92T/L145M/C321S/R134E/K138P) showed a 36.3-fold enhancement over the wild-type, and the second-round mutant M7 (N50S/K92T/L145M/C321S/R134E/K138P/M315G) achieved a 43.1-fold improvement. This highly efficient process demonstrates that integrating computational prediction with strategic library design can effectively reduce the screening efforts while maximizing the discovery of superior mutations.

Coevolution-based computational prediction efficiently enriches beneficial targets among a vast number of mutations, but predictions are not absolutely accurate. Among the 14 selected candidate mutations, most single-point mutations resulted in reduced activity, with significant improvements only observed for specific combinations. ΔE and effect strength scores reliably identify “functionally sensitive” sites, but uncertainty remains regarding the predicted direction of effect for single-point mutations. For instance, R134E and K138P each exerted a negative effect when introduced individually into the M4 backbone, whereas their rationally designed combination yielded strong positive synergy, underscoring how epistatic interactions within an optimized scaffold can reverse individual mutational effects. Rational combinations more effectively capture and leverages epistatic effects between residues, improving engineering success rates.

The success of this strategy is largely attributed to the abundant homologous sequences of TcCGT1, which provide the depth required for statistical power in coevolutionary analyses. For proteins with scarce homologous sequences, insufficient MSA depth leads to weak evolutionary coupling signals, which may reduce prediction accuracy. In the future, integrating coevolutionary constraints with physics-based molecular dynamics simulations or deep learning predictions is expected to address insufficient sequence data and enhance the method's generalizability.

Characterizing the enhanced catalytic performance

3.3

The engineered variants M4 and M7 exhibited considerable improvements in catalytic performance toward different substrates and thermal stability compared to the WT (Fig. 3A). Detailed parameters are provided in Table S6. With UDP-Glc as the donor, both variants displayed enhanced activity across structurally diverse flavonoid acceptors to generate C-, O-, and S-glycosides (Fig. 3B–S7-S20). Notably, M7 achieved a 43.1-fold increase in activity toward apigenin and a greater than 50.0-fold enhancement for kaempferol. In terms of synthesis capacity, M7 produced 3.69 ± 0.02 g of vitexin per liter of LB medium, which is a 62.1-fold increase over the wild-type (Fig. 3A). This result demonstrates that the integrated structure-coevolution engineering strategy not only enhanced the specific activity and stability but also substantially improved the total catalytic output of the enzyme.Fig. 3Enhanced catalytic performance and broadened substrate scope of engineered TcCGT1 variants. (A) Radar chart comparing multiple enzyme parameters from WT (black) to mutants M4 (blue) and M7 (red). Synthesis capacity means the maximum amount of vitexin produced by the enzyme expressed from 1 L of LB medium. (B) Glycosylation activity of WT, M4, and M7 toward different flavonoid acceptors (1-15, Table S3) using UDP-Glc as the sugar donor. The aglycone substrates (1–15) are: Apigenin; Luteolin; Chrysin; Oroxylin A; Kumatakenin B; Galangin; Kaempferol; Kaempferide; Naringenin; Liquiritigenin; 1,6-Dihydroxynaphthalene; Wogonin; Licoflavone C; (−)-Maackiain; 3,4-Dichlorobenzenethiol. (C) Glycosylation activity of WT and mutant M7 with different UDP-sugar donors. Error bars represent the standard deviation of three replicates. N.D., not detected.Fig. 3

The mutants also showed improved activity toward alternative sugar donors, although UDP-Glc remained the preferred donor. M7 exhibited 16.0- and 30.0-fold increases in catalytic efficiency with UDP-Gal and UDP-Xyl, respectively. No activity was detected with non-cognate donors, indicating preserved donor specificity boundaries (Fig. 3C). Complementing the enhanced activity was a substantial improvement in operational stability. The operational stability was extended from less than a day to over 10 days, accompanied by a 2.9 °C increase in melting temperature (Fig. S2). This improvement was corroborated by elevated melting temperatures rising from 57.7 °C for the wild-type to 58.7 °C for M4 and 60.6 °C for M7. The combination of broad substrate versatility, elevated activity, and outstanding stability renders these engineered enzymes highly promising for industrial biocatalysis, with profound green chemistry significance.

From the perspective of enzyme kinetic parameters, changes in Km, kcat, and kcat/Km are a direct quantitative reflection of the activity-stability trade-off (Fig. 3A–S3) [44,45]. The improvement in catalytic efficiency of the acceptor apigenin (kcat/Km) was primarily driven by a >6-fold increase in turnover number (kcat), despite a modest reduction in binding affinity (increased Km). In contrast, M7 achieved a >16-fold higher efficiency of the donor UDP-Glc through a combination of both enhanced affinity (lower Km) and accelerated catalysis (higher kcat).

These three parameters are closely related to the balance between enzyme conformational flexibility and structural stability [46]. The core requirement for catalytic activity is that the active site possesses appropriate flexibility to ensure substrate binding (resulting in decreased Km) and catalytic reaction progression (resulting in increased kcat). In contrast, improved structural stability often relies on enhanced overall conformational rigidity; improper regulation can easily lead to insufficient flexibility of the active site, thereby causing decreased kcat, increased Km, and ultimately reduced kcat/Km, forming a typical trade-off. Conversely, overemphasizing active site flexibility to improve kcat and reduce Km will compromise the overall structural stability of the enzyme, making it difficult to adapt to non-physiological conditions in industrial production. Only through precise regulation can the synergistic optimization of Km, kcat, and kcat/Km be achieved while enhancing structural stability and maintaining or optimizing the substrate-binding and catalytic conformations of the active site, thereby breaking the trade-off constraint. In our mutants, substrate affinity toward apigenin is slightly reduced in the engineered variants, but their elevated turnover rate compensates for this loss, leading to improved catalytic efficiency. This is a particularly advantageous feature in industrial settings where high substrate concentrations are common.

Structural basis for the improved catalytic performance of M7

3.4

Our integrated structural and dynamic analyses reveal that the superior performance of M7 originates from the synergistic effect of its seven mutations, which collectively break the classic activity-stability trade-off by coupling active-site-localized catalytic optimization with distal-site-mediated global conformational stabilization. The most direct structural evidence lies in the reconfigured substrate-binding mode (Fig. 4A). In the WT complex, apigenin forms key hydrogen bonds with the catalytic residue H24 [34]. In M7, active-site mutations (N50S/K92T/L145M/C321S) drive an approximately 120° rotation of apigenin, enabling it to form new hydrogen bonds with S321, F322, and D189, while UDP-Glc exhibits only a minor shift. We propose that this repositioning pre-organizes the substrate closer to the transition state, which directly explains the more than six-fold increase in kcat for apigenin.Fig. 4Structural basis for the improved performance of the M7 variant. (A) Locations of the seven mutations (active-site: blue spheres; distal-site: pink spheres) in the overall structure complexed with apigenin (mint) and UDP-Glc (yellow). Key residues and hydrogen bonds (yellow dashes) are shown in stick representation. The key residues of WT (blue) and M7 (orange) are shown in stick representation. (B) Predicted changes in the substrate-binding cavity. (C) Analysis of molecular dynamics simulations of the WT and M7 complexes. Root-mean-square deviation (RMSD) of the protein backbone over the simulation time course. (D) Root-mean-square fluctuation (RMSF) per residue for the WT and M7 proteins. Gray shading indicates regions discussed in the text.Fig. 4

This substrate repositioning is facilitated by systematic reshaping of the active-site architecture induced by the active-site mutations [45]. CavityPlus calculations demonstrated changes in the cavity located in the interdomain cleft, a typical feature of GTB family transferases [34]. Specifically, the cavity volume decreased from 791 to 742 Å^3^ and the surface area from 821 to 785 Å^2^ (Fig. 4B). This more compact cavity enhances catalytic efficiency by improving substrate binding specificity and transition-state stabilization [47]. The observation is consistent with the pre-organized substrate conformation and increased kcat of M7. Concurrently, access channel diameters increased from 5.5 to 7.0 Å and from 6.4 to 10.1 Å. This combination of a tighter binding pocket and widened channels optimizes substrate ingress and product egress, supporting the higher turnover rate. The tighter pocket optimizes transition-state stabilization while the widened channels facilitate substrate ingress and product egress, which is reflected by the more than six-fold increase in kcat.

The improved catalytic efficiency is firmly anchored in enhanced global conformational stability driven by distal-site mutations (R134E/K138P/M315G) [48]. Molecular dynamics simulations showed clear differences in structural stability between M7 and WT. M7 remained stable with a value of 2.12 ± 0.13 Å, whereas the WT showed greater flexibility with a higher and more variable RMSD of 2.45 ± 0.21 Å (Fig. 4C). Consistent with this finding, root-mean-square fluctuation (RMSF) analysis identified increased flexibility in specific regions of WT, including residues 48-61, 167-169, 250-270, and 310-320, which are rigidified in M7 (Fig. 4D). This rigidification mediated by distal mutations is crucial for maintaining the precise catalytic geometry achieved by the active-site mutations. It prevents the loss of the optimized substrate-binding and transition-state-stabilizing conformations under physiological or reaction conditions.

Distal-surface mutations contribute to this stability by modulating the protein's electrostatic landscape [[49], [50], [51]]. Specifically, these mutations replaced positive and neutral surface residues with negatively charged or alternative neutral amino acids, effectively reducing the surface electrostatic potential energy (Fig. S4). This charge redistribution likely mitigates inherent repulsive interactions, lowering the overall free energy of the folded state and thereby enhancing both structural integrity and conformational rigidity [[52], [53], [54]].

In summary, M7 represents a synergistic state where local catalytic prowess from active-site mutations is reinforced by global stabilization from distal-site mutations. Active-site mutations reconfigure the substrate-binding mode and reshape the active-site cavity to enhance catalytic efficiency, which is reflected by the kinetic parameters. Distal-site mutations, on the other hand, rigidify the global structure and optimize the electrostatic landscape to maintain the precise catalytic geometry. This coupling effectively breaks the classic activity-stability trade-off, explaining M7's simultaneous improvements in catalytic efficiency, thermostability, and substrate specificity.

Our structure-coevolution dual-guided engineering implements a synergistic optimization paradigm integrating active-site modification and distal functionally coupled residue engineering: through EVcouplings-based coevolutionary analysis, we directly identified distal residues (e.g., R134E/K138P/M315G) that exhibit evolutionary constraint relationships with the active site. These residues are not merely stability-enhancing sites; instead, they enhance the catalytic efficiency of the active site via conformational regulation, thereby achieving synergy between improved catalytic activity and enhanced thermostability.

De novo production of vitexin in Yarrowia lipolytica

3.5

We systematically engineered the flavonoid biosynthetic pathway in Y. lipolytica by modular optimization of three modules: precursor supply (shikimate pathway), core pathway (flavonoid synthesis), and glycosylation module (engineered CGT). Given the pivotal role of naringenin as the precursor, we first constructed a naringenin-producing chassis. The initial strain NAG01 was developed by knocking out endogenous gene dga1 and dga2 to redirect carbon flux from lipid storage toward flavonoid biosynthesis [55]. We also integrated fjtal (from Flavobacterium johnsoniae) and pc4cl (from Petroselinum crispum) to establish the phenylpropanoid pathway from tyrosine, along with phchs (from Petunia hybrida) and mschi (from Medicago sativa) to enable naringenin formation via malonyl-CoA condensation in NAG02 (Fig. 5A). NAG01 produced 43.6 ± 1.4 mg/L naringenin, confirming pathway functionality while clearly revealing limitations in aromatic amino acid supply and malonyl-CoA supply, the primary precursor bottleneck for flavonoid synthesis.Fig. 5De novo biosynthesis of vitexin in engineered Yarrowia lipolytica. (A) Schematic of the constructed biosynthetic pathway. Overexpressed heterologous and endogenous genes are highlighted; deleted genes are marked with a cross. Key metabolic nodes (e.g., UDP-glucose, malonyl-CoA) and precursors are indicated. (B) Naringenin (vitexin precursor) production in pathway-engineered strains. (C) Vitexin production by different glycosyltransferases (WT or M7) and in strains with or without the enhanced endogenous UDP-glucose biosynthetic pathway. HPLC profiles of fermentation broths directly compare the titers of vitexin and the major byproduct, naringenin-C-glycoside. (D) Optimization of the final producer strain. Restoring lipid homeostasis by reintroducing endogenous dga1 or dga2 to improve growth and production. Enhancing pathway flux by overexpressing key bottleneck genes. “+” indicates that the gene has been modified; “−” indicates that the gene has not been modified. Error bars represent the standard deviation of three independent cultures initiated from single colonies. Statistical analysis was performed by one-way ANOVA (∗p < 0.05, ∗∗p < 0.01, ∗∗∗p < 0.001).Fig. 5

To address this precursor supply bottleneck, we implemented a combinatorial optimization strategy: deregulating the shikimate pathway with feedback-resistant mutant alleles aro4^K229L^ and aro7^G141S^, constructing a tyrosine-phenylalanine shunt via atcpr, atpal, and atc4h to enhance coumarate flux; and overexpressing endogenous acc1 to increase malonyl-CoA availability [56]. Through combinatorial optimization, the resulting strain NAG07 achieved a naringenin titer of 117.5 ± 5.0 mg/L, a 169% increase over NAG01, validating the effectiveness of targeted precursor pathway engineering. (Fig. 5B).

We next evaluated the glycosylation capability of the WT versus the engineered variant M7 within this chassis, focusing on the cofactor supply and enzyme catalytic efficiency bottleneck. Strain VIT02 (M7) produced 18.9 ± 4.2 mg/L vitexin, a 2.4-fold higher titer than that of control strain VIT01 (WT), confirming M7's enhanced in vivo activity. To further amplify this advantage by addressing UDP-glucose limitation, we enhanced endogenous UDP-glucose supply by overexpressing two endogenous genes: ugp1 (encoding UDP-glucose pyrophosphorylase) and pgm2 (encoding phosphoglucomutase) [57]. VIT04 (M7 with enhanced endogenous UDP-glucose supply) then reached a titer of 45.8 ± 3.2 mg/L, representing a 6.3-fold increase over VIT01 (WT) and a 2.0-fold increase over VIT03 (WT with enhanced endogenous UDP-glucose supply) (Fig. 5C). These results demonstrate that M7 not only possesses higher intrinsic activity but also efficiently utilizes enhanced glycosyl-donor availability to significantly improve vitexin production (Fig. 5C).

Beyond higher activity, M7 exhibited superior pathway specificity by strongly favoring apigenin over the intermediate naringenin, solving the byproduct accumulation bottleneck. The sole major by-product, naringenin-C-glycoside, is derived from the glycosylation of the intermediate naringenin. Under elevated UDP-glucose endogenous supply, strain VIT03 (WT with enhanced endogenous UDP-glucose supply) accumulated 96.0 ± 4.0 mg/L of the by-product naringenin-C-glycoside, whereas strain VIT04 (M7 with enhanced endogenous UDP-glucose supply) produced only 26.7 ± 2.3 mg/L—a 3.6-fold reduction. This specificity-enhancing effect demonstrates that M7 has a stronger substrate preference for apigenin, which enables it to reduce futile side reactions even when glycosyl donor pools are abundant. Collectively, the M7 variant simultaneously maximizes vitexin productivity and lowers downstream purification costs, thus serving as a pivotal catalyst for the economical microbial synthesis of vitexin and related C-glycosyl flavonoids.

To resolve the growth-production trade-off bottleneck caused by the dga1/dga2 knockout, which induced metabolic burden and limited biomassx we reintroduced these corresponding lipid synthesis genes. This knockout not only suppressed lipid droplet biosynthesis but also triggered metabolic burden. Reintroducing these lipid synthesis genes led to a distinct asymmetric effect on flavonoid production: VIT06 (restoring dga2 expression) markedly increased vitexin titer to 220.5 ± 8.4 mg/L, while VIT05 (restoring dga1 alone) or VIT07 (restoring both genes together) resulted in lower yields (Fig. 5D). This difference reflects their distinct catalytic roles: dga2 assembles small, highly dynamic lipid droplets with rapid turnover. Its moderate consumption of malonyl-CoA/acetyl-CoA helps restore membrane lipid homeostasis and redox balance without excessively draining the precursor pool for flavonoid synthesis. Therefore, lipid engineering targeting dga2 represents an optimal strategy to achieve metabolic balance, alleviating stress while preserving precursor availability, thereby addressing the growth-production trade-off in lipid-metabolically engineered chassis.

Finally, to tackle the pathway flux bottleneck, we enhaced pathway flux by increasing the copy numbers of pcfns, pc4cl, and phchs. The combinatorial amplification of these genes in strain VIT10 resulted in the highest vitexin titer of 361.0 ± 7.5 mg/L, confirming that elevated expression of key enzymes acts synergistically to improve yield (Fig. 5D).

Enhancing vitexin production by fed-batch fermentation

3.6

Fed-batch fermentation was scaled up in a 5 L bioreactor to further optimize vitexin production. Engineered strain VIT10 was cultured under the previously optimized parameter conditions (Fig. S5). However, fermentation using an inorganic nitrogen source throughout the process exhibited inherent limitations. Cell growth was relatively slow in the first 96 h, with a maximum OD_600_ of only about 120, and the vitexin titer remained low at 847.8 mg/L (Fig. 6B). We hypothesized that although inorganic nitrogen enables precise control of the C/N ratio and channels metabolic flux toward product synthesis, its slow assimilation rate during the initial fermentation phase fails to meet the nitrogen requirements for rapid cell proliferation. This limitation restricts biomass accumulation and indirectly compromises product synthesis.Fig. 6De novo synthesis of vitexin in a 5 L fermenter by strain VIT10. (A) Vitexin was produced by fed-batch fermentation with inorganic nitrogen sources. (B) Vitexin was produced by fed-batch fermentation with YPD as the initial nitrogen source, and supplemented with inorganic nitrogen sources. Black arrows denote glucose feeding to maintain concentration, and blue arrow indicates the switch to inorganic nitrogen sources. The substrate feeding was performed in a continuous and automated mode to maintaining the residual glucose below 1 g/L.Fig. 6

To mitigate sluggish cell growth, a staged nitrogen feeding strategy was developed. Organic nitrogen was used during the exponential growth phase to promote biomass accumulation, followed by a switch to inorganic nitrogen during the production phase. This hybrid strategy yielded significant improvements: the cells utilized organic nitrogen for rapid proliferation, reaching an OD_600_ of 149.1 within 96 h (Fig. 6A). Subsequent feeding with inorganic nitrogen transitioned the culture into a stable product synthesis phase. Ultimately, we achieved a titer of 2.14 g/L at 168 h, a yield of 10.84 mg/g glucose, and a productivity of 12.74 mg/L/h. The obtained yield is well within the reasonable range of reported values for microbial synthesis of flavonoid compounds, as summarized in related studies [41,58]. Notably, apigenin accumulation remained consistently low (<198.7 mg/L), indicating high metabolic flux and conversion efficiency throughout the downstream pathway from l-tyrosine to vitexin. The titer of naringenin-C-glucoside was maintained below 50 mg/L, substantially lower than that in shake flask cultures. This reduction is likely attributed to the increased copy number of key genes, which redirected metabolic flux toward vitexin synthesis and minimized the accumulation of this upstream byproduct.

Integration of the optimized parameters with the engineered staged nitrogen feeding strategy in the bioreactor successfully overcame the limitations of the single nitrogen source approach. This integrated scheme balances the distinct nutritional demands of cell growth and product synthesis: organic nitrogen ensures rapid cell proliferation in the early stage, while precise control of inorganic nitrogen in the later stage guides efficient substrate conversion into the target product. Ultimately, this strategy achieved a high vitexin titer of 2.14 g/L via de novo synthesis in engineered microbes using low-cost carbon sources. While this titer is lower than the state-of-the-art production in optimized chassis [59], it convincingly validates the high in vivo performance of our engineered CGT and establishes a foundational platform for synthesizing vitexin in Y. lipolytica. The structure-coevolution dual-guided strategy, as the core of this work, provides a generalizable framework for overcoming the activity-stability trade-off in glycosyltransferase engineering.

Conclusion

4

In this study, we constructed a full-chain synthetic biology platform for the sustainable de novo biosynthesis of vitexin—a high-value flavonoid C-glycoside—by integrating enzyme engineering, metabolic pathway optimization, and fermentation. We developed a structure-coevolution dual-guided engineering strategy to address the long-standing limitation of the activity–stability trade-off. This strategy not only provides a robust and generalizable framework for CGT engineering but also highlights the potential of integrated strategies to advance sustainable biomanufacturing. To translate this enzymatic breakthrough into practical bioproduction, we systematically engineered Yarrowia lipolytica as a chassis strain via modular optimization. The resulting microbial cell factory enables vitexin production from glucose, offering clear green and sustainable prospects with environmental advantages. Future work will focus on two key directions: first, resolving the high-resolution structures of these mutants to elucidate the precise conformational changes underlying their improved performance; second, extending this strategy to other GT families to explore its applicability for the biosynthesis of diverse glycosylated natural products.

CRediT authorship contribution statement

Chu-Qi Shi: Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Yi-Jing Chen: Validation, Investigation, Data curation. Wen-Li Yang: Validation, Investigation, Data curation. Shuang Zheng: Validation, Investigation, Data curation. Rong Cai: Supervision, Resources, Methodology, Conceptualization. Jin Hou: Supervision, Resources, Methodology, Conceptualization. De-Qing Wang: Writing – review & editing, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation. Ju-Zheng Sheng: Writing – review & editing, Supervision, Project administration, Methodology, Funding acquisition, Formal analysis, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.